1 / 27

Memory Allocator Attack and Defense

Memory Allocator Attack and Defense. Richard Johnson richardj@microsoft.com switech@microsoft.com. Dynamic Memory Management. The memory manager is responsible for tracking a program’s dynamic data storage.

flower
Download Presentation

Memory Allocator Attack and Defense

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Memory Allocator Attack and Defense Richard Johnson richardj@microsoft.com switech@microsoft.com

  2. Dynamic Memory Management • The memory manager is responsible for tracking a program’s dynamic data storage. • Unlike stacks which work based upon a simple FIFO/LIFO concepts, heaps require management routines to track the location of free and allocated memory chunks

  3. Dynamic Memory Management • What approaches to dynamic memory management have been developed? • What are the security profiles of memory managers used in mainstream OS’s today? • What is the impact of security research on memory manager design?

  4. Dynamic Memory Management • Today we will consider the following OS’s and their memory allocators: • Windows • Linux • Apple OS X • OpenBSD

  5. Dynamic Memory Management • Today we will consider the following OS’s and their memory allocators: • Windows • Windows Heap Manager • Rockall Allocator • Linux • Doug Lea Malloc • Apple OS X • Poul-Henning Kamp Malloc • OpenBSD • OpenBSDMalloc

  6. What’s the Difference? • The primary difference between the memory managers is how they track free buffers • We will split them into systems that inline management data on each chunk and those that do not • Management data inlined in the heap is susceptible to modification when a memory corruption occurs

  7. What’s the Difference? • Heaps with inlined management structs expose user APIs that walk linked lists of buffers to locate the appropriate buffer • Doug Lea • Windows Heap Manager • Heaps without inlined management data try to take advantage of kernel-supplied memory management APIs and utilize array indexing to locate buffers • Poul-Henning Kamp • OpenBSDMalloc • Rockall

  8. Security Research on Heap Allocators • Offensive security researchers focus on adding reliability to exploitation methods or finding new ways to manipulate management routines to gain controllable memory corruption • Defensive security researchers aim to mitigate known attacks or (rarely) attempt new heap manager designs

  9. Security Research on Heap Allocators • dlmalloc • 2001 Michel "MaXX" Kaempf / Anonymous • 2005 Phantasmal Phantasmagoria • Windows Heap • 2002 David Litchfield • 2004 Matt Conover / OdedHorovitz • 2005 SecurityPatrol

  10. Security Research on Heap Allocators • PHKMalloc • 2005 Yves Younan et al • OpenBSDMalloc • 2006 Ben Hawkes

  11. Heaps with inline data • Basic mechanics: • A region of memory is allocated to contain buffers • An array of doubly linked lists tracking free buffers in multiples of a fixed size (usually 8) is created • On allocation a free chunk is unlinked from the doubly linked list and the address is returned to the program • On free, a 8 byte header is written to the beginning of a buffer and the chunk is added back to the list • When two free buffers are adjacent they will be merged into one larger chunk of free memory • Lookaside lists*

  12. Heaps with inline data • Attacks • Unlink • Free buffer is removed from doubly linked list with corrupted forward and backward pointers • Attacker writes 4 bytes of controlled data to a controlled location • Coalesce • Manipulating the flag indicating whether the previous chunk is in use can be used with a fake chunk header to cause a 4 byte write to a controlled location • Lookaside list • The head of a lookaside list can be overwritten to later return a controlled address to the next allocation of that size

  13. Heaps with inline data • Unlink Attack • Scenario: Heap-based buffer overflow allows for writing into adjacent free heap block • Attack: Overwrite FLINK and BLINK values and wait for next allocation • Result: Allows one or more 4-byte writes to controlled locations FREE HEAP BLOCK_HEAP_ENTRY +0x000 Size +0x002 PreviousSize +0x004 SmallTagIndex +0x005 Flags +0x006 UnusedBytes +0x007 SegmentIndex _LIST_ENTRY +0x000 Flink +0x004 Blink movdwordptr [ecx],eaxmovdwordptr [eax+4],ecx EAX = Flink, ECX = Blink

  14. Heaps with inline data • Lookaside Attack • Scenario: Heap-based buffer overflow allows for control of lookaside list management structure • Attack: First heap overwrite takes control of Flink value in a free chunk with a lookaside list entryAllocation of the corrupted chunk puts the corrupt Flink value into the lookaside listNext HeapAlloc() of the same sized chunk will return the corrupted pointer • Result: Returns corrupted pointer from the next allocation from the lookaside list which allows for arbitrary length overwrites

  15. Heaps without inline data • Basic mechanics: • Relies on and optimized for kernel provided virtual memory management system • Heap manager tracks allocated pages, allocated chunks and free pages in a series of directories • All chunks in a page are typically of the same size • Adjacent free pages are coalesced

  16. Heaps without inline data • Attacks • free() • Control of a pointer passed to free can be abused to free memory that contains one of the heap management structures. • pginfo / pgfree • Manipulate the value returned by an allocation

  17. Heaps without inline data • free() attack • Scenario: Heap-based buffer overflow allows for control of pointers later passed to free() • Attack: Free pages with control structures on them • Result: Later allocations will eventually return the page with the control structures and allow for further exploitation

  18. Heaps without inline data • pginfoattack • Scenario: Heap-based buffer overflow allows for control of the pginfostructure leading to arbitrary memory corruption • Attack: Heap overflow allows for modification of the pginfo->free page pointer. Overwrite bits array to make pages seem free • Result: Allocation requests walk the structs to find the appropriate sized buffers so returning corrupted pointer allows for writes to arbitrary locations. PGFREE structpgfree { structpgfree *next; structpgfree *prev; // free pages void *page; // base page dir void *pdir; // bytes free size_t size; }; PGINFO structpginfo{ structpginfo*next; void *page; ushortsize; ushortshift; ushortfree; ushorttotal; uint bits[]; };

  19. Heap Allocator Defense • dlmalloc • glibc added safe unlinking • Windows Heap • Safe unlinking • Checksum for size and flags • XOR size, flags, checksum, and prevsize fields • Lookaside list replaced by LFH in Vista

  20. Heap Allocator Defense • phkmalloc • Nada • OpenBSDmalloc • Nada • System defenses such as ASLR and NX also apply but are not part of the heap manager’s architecture

  21. So what’s next?

  22. Windows Kernel Pool Manager “The Month of Kernel Bugs is a serious wake-up call about the vulnerability of the most fundamental element of the operating system. Begin preparing now for more, and more damaging, attacks against the OS kernel.” Rich Mogul – Gartner Nov. 2006http://www.gartner.com/resources/144700/144700/learn_from_month_of_kernel_b_144700.pdf

  23. Windows Kernel Pool Manager • 2005 SoBeIt “How to exploit Windows kernel memory pool” • Basic unlink() technique applies to the kernel pool

  24. Windows Kernel Pool Manager • Pools are managed by a pool descriptor, chunks are managed by a pool chunk header lkd> dt -v -r nt!POOL_DESCRIPTOR struct _POOL_DESCRIPTOR, 14 elements, 0x1034 bytes +0x000 PoolType : Enum _POOL_TYPE +0x004 PoolIndex : Uint4B +0x008 RunningAllocs : Int4B +0x00c RunningDeAllocs : Int4B +0x010 TotalPages : Int4B +0x014 TotalBigPages : Int4B +0x018 Threshold : Uint4B +0x01c LockAddress : Ptr32 to Void +0x020 PendingFrees : Ptr32 to Ptr32 to Void +0x024 ThreadsProcessingDeferrals : Int4B +0x028 PendingFreeDepth : Int4B +0x02c TotalBytes : Uint4B +0x030 Spare0 : Uint4B +0x034 ListHeads : [512] struct _LIST_ENTRY, 2 elements, 0x8 bytes +0x000 Flink : Ptr32 to struct _LIST_ENTRY, 2 elements, 0x8 bytes +0x000 Flink : Ptr32 to struct _LIST_ENTRY, 2 elements, 0x8 bytes +0x004 Blink : Ptr32 to struct _LIST_ENTRY, 2 elements, 0x8 bytes +0x004 Blink : Ptr32 to struct _LIST_ENTRY, 2 elements, 0x8 bytes +0x000 Flink : Ptr32 to struct _LIST_ENTRY, 2 elements, 0x8 bytes +0x004 Blink : Ptr32 to struct _LIST_ENTRY, 2 elements, 0x8 bytes lkd> dt -v -r nt!POOL_HEADER struct _POOL_HEADER, 8 elements, 0x8 bytes +0x000 PreviousSize : Bitfield Pos 0, 9 Bits +0x000 PoolIndex : Bitfield Pos 9, 7 Bits +0x002 BlockSize : Bitfield Pos 0, 9 Bits +0x002 PoolType : Bitfield Pos 9, 7 Bits +0x000 Ulong1 : Uint4B +0x004 PoolTag : Uint4B +0x004 AllocatorBackTraceIndex : Uint2B +0x006 PoolTagHash : Uint2B

  25. Windows Kernel Pool Manager • The good news • We’re active researching how to add appropriate mitigations to the kernel memory management code • The bad news • Unlike user heaps, the kernel pool is globally managed • There aren’t any free bytes to use for checksums and cookies • Performance and compatibility concerns sometimes trump security

  26. Windows Kernel Pool Manager • You can help. Contact us at switech@microsoft.com if you are interested in this research and want your ideas heard!

  27. Questions?

More Related